Prior knowledge scaffolds acquisition of new memories but only sleep supports integration

Our understanding of the world informs how we encode new experiences, and new experiences in turn shape our understanding. How do we balance these critical reciprocal processes over time? Existing theories posit that interleaved reactivation of old and new information during sleep may play an important role. To evaluate this idea, we developed a paradigm where participants first learned about objects from novel, hierarchically structured categories and subsequently encountered new exemplars from those categories. In four initial experiments, participants completed both learning phases in one session, allowing us to examine whether participants rapidly acquire and integrate new structured knowledge without sleep. We found consistent evidence that participants quickly learned the category structure and used their knowledge to scaffold learning of the new exemplars. However, the newly learned exemplars were not integrated into existing knowledge; rather, the representations of learned exemplars were strongly shaped by the temporal context (i.e., the learning phase) in which they were encoded. In a subsequent experiment, we tested the hypothesis that in order to integrate new information, the replay of new memories must be interleaved with the replay of old memories during sleep. We employed a real-time EEG approach to cue the reactivation of specific memories during a nap, and found that interleaved reactivation indeed promoted integration across contexts. A neural network model implementing context suppression during sleep recapitulated these results. Together, these results inform our understanding of how sleep abstracts memories away from their episodic context to facilitate integration across related memories.